| Benchmark | Claude 3.7 Sonnet | Claude 4 Opus |
|---|---|---|
| AA Intelligence | 27.6 | 31.7 |
| AA-LCR | 62.3 | 40 |
| aider_polyglot | 64.9 | 70.7 |
| AIME 2025 | 54.8 | 75.5 |
| AIR-Bench 2024 | 81.8 | 85.7 |
| anthropic_red_team | 99.7 | 98.9 |
| arc_agi_1 | 13.6 | 27 |
| ARC-AGI-2 | 0 | 8.6 |
| arena_vision | 1195 | 1207 |
| Artificial Analysis Coding Index | 36.4 | 34 |
| bbq | 92.1 | 99.3 |
| Fortress | 38 | 27.6 |
| GPQA Diamond | 84.8 | 79.6 |
| harmbench | 84.3 | 91.7 |
| HLE | 9.7 | 12.3 |
| hmmt_feb_2025 | 31.7 | 60 |
| IFBench | 48.3 | 53.7 |
| ifeval | 93.2 | 87.4 |
| math_500_em | 96.2 | 98.2 |
| metr_hcast | 54.2 | 61.7 |
| metr_re_bench | 0 | 20 |
| metr_swaa | 98.7 | 99.8 |
| mmmlu | 86.1 | 88.8 |
| MMMU | 75 | 73.7 |
| multichallenge | 51.6 | 58.6 |
| MultiNRC | 27.8 | 33.9 |
| scicode | 40.3 | 40.9 |
| SEAL VISTA | 48.2 | 47 |
| simple_safety_tests | 100 | 100 |
| simplebench | 46.4 | 58.8 |
| swe_bench_bash | 52.8 | 67.6 |
| SWE-bench Verified | 70.3 | 79.4 |
| TAU-bench (airline) | 58.4 | 59.6 |
| TAU-bench (retail) | 81.2 | 81.4 |
| Terminal-Bench | 35.2 | 43.2 |
| Terminal-Bench Hard | 21.2 | 31.1 |
| xstest | 96.4 | 97 |
| τ²-Bench Telecom (AA run) | 54.7 | 73.4 |
Best tracked score per model per benchmark (default configuration; source-attributed). ↓ marks lower-is-better metrics. Open either model for its full surface, provenance and pricing.